Speed up CI and the next bottleneck just moves to deploy and rollback, which do not scale the same way.
Why not? A good chunk of software engineering problems can be solved if we re-structured our definitions. Even just a year ago, people did not believe that a good chunk of prod-running code would be completely AI generated. Now we're at the point where some companies try to split "what code has to be PR-reviewed, and what doesn't have to be" to remove that bottleneck and so on.